How Well do LLMs Compress Their Own Chain-of-Thought? A Token Complexity Approach

Lee, Ayeong, Che, Ethan, Peng, Tianyi · arXiv · 2025